Seminars and Events
Do We Need Large Language Models for Time Series?
Event Details
Speaker: Vinayak Gupta, Lawrence Livermore National Laboratory
Location: Virtual Only via Zoom
Zoom Details:
https://usc.zoom.us/j/98248194762?pwd=KCPIsauraEJDFnw102leuBjxehbbiM.1
Meeting ID : 982 4819 4762
Passcode: 470845
Register in advance for this webinar:
https://usc.zoom.us/webinar/register/WN_78–B06ZRNub3zx6WKvfmg
After registering, you will receive a confirmation email containing information about joining the webinar.
Visit links below to subscribe and for details on upcoming seminars:
https://www.isi.edu/isi-seminar-series/
Recent large language models (LLMs) have only shown potential for reasoning with text and image data. We explore this reasoning ability with one of the most important data formats: time-series. Capturing the sequential nature of time-series data is crucial to power applications in finance and healthcare. This talk presents a first-of-its-kind benchmark that focuses on truly understanding time-series data and goes beyond the existing evaluations. Additionally, we will discuss the notable limitations of existing works claiming that LLMs can perform forecasting. Our analysis across such models finds that simply removing the LLMs or replacing them with a basic attention layer improved results in most cases, and also led to better scalable solutions.
Speaker Bio
Vinayak Gupta is a researcher in the AI Research Group at the Lawrence Livermore National Laboratory. Prior to this, he was a postdoctoral scholar at the University of Washington, Seattle, and an AI Scientist at IBM Research. His research focuses on mining large-scale time-series data, and more recently, he has been working on leveraging LLMs to jointly understand text+time-series. He received his PhD from the Indian Institute of Technology, Delhi in 2022. He was a runner-up in the AI Gamechangers of India and was featured as an AI expert in India AI, the AI initiative of the Government of India.
Host: Abel Salinas, POC: Pete Zamar
If speaker approves to be recorded for this AI Seminar talk, it will be posted on our USC/ISI YouTube page within 1-2 business days: https://www.youtube.com/user/USCISI.